Prediction of full-scale filtration plant performance using artificial neural networks based on principal component analysis

Abstract

To obtain standard water quality is one of the most crucial issues must be discussed. To get higher water quality, the separation and purification processes must be applied. In this study, 44 water quality parameters were monitored between May 2018 and February 2019 in order to evaluate the efficiency of a full-scale filtration plant which uses particulate-, micro- and ultrafiltration processes as a pre-treatment and applied reverse osmosis as the post-treatment. A crucial research question of this study was thus whether the Ultrafiltration (UF) and Reverse Osmosis (RO) systems performance about the elimination of monitored parameters. The most striking results from the monitoring study reveal that the pre-treatment processes are not suitable for separation of uni/polyvalent ions. The hypothesis that will be tested for both systems; the UF process was efficient for microbial and nitrogen parameters and the RO was efficient for separation of anions and metals that are identified priority hazardous substance. Then, the purification performance of a filtration plant was evaluated using the Artificial Neural Network (ANN) model based on the Principal Component Analysis (PCA) that used to reduce the number of input water parameters. PCA components were used as input in the model and according to the results of Pearson Correlation Analysis, the conductivity parameter which was directly or indirectly related with almost all parameters was used as output. Consistency of created ANN model with real data was 98.758% and mean square errors was 0.00293.